The Sentinel Initiative mandated by the Food and Drug Administration will lead to an enormous number of studies being planned post-market that will require analyzing and combining data from several different studies. The proposed project will address this challenge through developing new and flexible methods for meta-analysis using a variety of models, including models for binary and discrete data, models for longitudinal data, and models for time-to-event data. A related issue that will also be addressed is design, sample size, and power considerations using these types of meta-analytic models. Such models and data collected post-market can be quite useful in designing future clinical studies such as non-inferiority, equivalence, and superiority cancer clinical trials. The proposed project will also develop methods for metaanalytic studies of diagnostic tests to facilitate evidence-based medicine. We will also create flexible and robust methodology for accurately comparing rare adverse event rates in cancer for different drugs and for determining how those rates are affected by important prognostic factors. The proposed project will also explore statistical methods for the analysis of large cancer data sets for calibrating treatment dose in the presence of potentially conflicting factors, such as length and quality of life and economic costs. We will explore these tradeoffs rigorously, using a utility based approach traditionally employed in the analysis of health policy at the population level. The proposed statistical methodology will be broadly applicable to complex, large scale, data sets arising in phase III clinical trials and post-marketing studies.